Multifunction test point allocation method and system based on adaptive switch matrix

Abstract

The application provides a multifunctional test point allocation method and system based on an adaptive switch matrix, relates to the technical field of electronic testing, and realizes dynamic connection of physical probes and logical test points by acquiring integrated circuit topology information to construct a switch matrix; the same probe is allocated to different test points in different time periods by using time division multiplexing technology, test response signals are collected and deviation values are calculated; a fault point is determined based on the deviation, the influence degree of the fault point on surrounding nodes is analyzed, and a test priority level is obtained; the test timing is dynamically adjusted according to the priority level, high-priority nodes are preferentially tested, the switch state is controlled to realize intelligent allocation of test resources, and the test efficiency and fault coverage are improved.

Description

Technical Field

[0001] This invention relates to electronic testing technology, and more particularly to a multifunctional test point allocation method and system based on an adaptive switch matrix. Background Technology

[0002] The field of integrated circuit testing has long faced the contradiction between limited physical probe resources and a large number of test points. Current testing methods typically employ a fixed probe allocation strategy, resulting in low testing efficiency and insufficient coverage, making it difficult to meet the comprehensive testing needs of complex integrated circuits. In existing technologies, the relationship between probes and test points is mostly a one-to-one static connection, which cannot dynamically adjust the testing strategy based on faults discovered during the testing process, leading to a waste of testing resources.

[0003] Existing methods lack in-depth analysis of the correlation between test points and fault propagation characteristics, making it difficult to effectively identify fault root causes and accurately assess affected areas. In high-density integrated circuit testing, limited probe resources cannot simultaneously cover all test points, while static polling methods are inefficient and struggle to target faulty areas. Current technologies also struggle to comprehensively consider multi-dimensional signal characteristics in the time and frequency domains for fault localization and lack adaptive optimization mechanisms, making it difficult for the testing process to self-adjust based on real-time test results. This hinders the efficient use of limited physical probe resources to achieve optimal test coverage and fault detection accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a multifunctional test point allocation method and system based on an adaptive switch matrix, which can solve the problems in existing technologies.

[0005] A first aspect of this invention provides a multifunctional test point allocation method based on an adaptive switching matrix, comprising:

[0006] Obtain the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes circuit connection relationships;

[0007] An adaptive switch matrix is ​​constructed based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switch units in different time periods, thereby generating a test timing schedule.

[0008] The test response signals of the logic test points are collected, and the deviation value of the test response signals relative to the expected signals is calculated. Fault test points are determined based on the deviation value. The degree of influence on test nodes connected to the fault test points is analyzed based on the circuit connection relationship to obtain the test priority level of the test nodes. The test timing schedule is adjusted according to the test priority level, and the physical probes are preferentially allocated to test nodes with higher priority levels to obtain an updated test timing schedule. The switching state of each controllable switch unit is controlled to achieve dynamic allocation of test points.

[0009] In one alternative implementation,

[0010] The steps for generating a test timing schedule by controlling the switching state of a controllable switching unit to allocate the same physical probe to different logic test points at different time periods include:

[0011] Obtain the feature vectors of the logical test points, calculate the correlation matrix between the test points based on the feature vectors of the logical test points, construct the probe allocation state space based on the correlation matrix, the probe allocation state space includes probe allocation state, resource constraints and test quality indicators; construct the probe allocation action space, the probe allocation action space includes probe number, test point number and allocation time period.

[0012] A reward function is constructed based on the probe allocation state space and the probe allocation action space. The reward function includes a test coverage reward, a test quality reward, and a constraint violation penalty. The probe allocation strategy is iteratively optimized based on the reward function. The probe allocation strategy satisfies the single-moment probe allocation constraint, the test point correlation constraint, and the minimum test time constraint.

[0013] Collect response signals from test points and calculate test quality assessment values. When the test quality assessment value is lower than a preset assessment threshold, update the weight parameters in the reward function based on the gradient descent method and re-execute the iterative optimization of the probe allocation strategy. Generate a test timing schedule table based on the optimized probe allocation strategy.

[0014] In one alternative implementation,

[0015] The probe allocation strategy is iteratively optimized based on the reward function. The steps of the probe allocation strategy satisfying the single-time probe allocation constraint, the test point correlation constraint, and the minimum test time constraint include:

[0016] A binary decision variable is established for the probe allocation constraint at a single time step. The binary decision variable represents the allocation relationship between the probe and the test point at each time step. The degree of violation of the correlation constraint of the test point is calculated based on the binary decision variable. The degree of violation of each constraint is normalized to obtain the normalized constraint violation degree. The constraint weight is dynamically updated based on the normalized constraint violation degree. The update of the constraint weight is achieved by multiplying the learning rate and the normalized constraint violation degree. A comprehensive evaluation index of constraint satisfaction is calculated based on the constraint weight.

[0017] A neighborhood structure for probe allocation schemes is constructed, which includes a probe allocation time adjustment scheme and a probe and test point redistribution scheme. Historical schemes are recorded to form a taboo table, and the probe allocation scheme is locally optimized based on the taboo table.

[0018] The optimization convergence is calculated based on the comprehensive evaluation index, and the optimization step size is dynamically adjusted based on the optimization convergence. The comprehensive evaluation index is used as the optimization basis for the reward function.

[0019] Calculate the variance of the probe allocation scheme after multiple consecutive optimizations. When the variance is less than a preset variance threshold or the number of consecutive optimizations reaches a preset number of iterations, determine the optimization result of the probe allocation strategy.

[0020] In one alternative implementation,

[0021] The steps for calculating the deviation of the test response signal from the expected signal include:

[0022] The test response signal is filtered to obtain a filtered signal. The filtered signal is divided into signal segments according to a preset time window. An expected signal template is established based on a standard waveform. The root mean square error between the signal segment and the expected signal template is calculated to obtain the time domain deviation. The spectrum between the signal segment and the expected signal template is calculated to obtain the frequency domain deviation. The delay deviation is calculated based on the cross-correlation function between the signal segment and the expected signal template.

[0023] The time-domain deviation, frequency-domain deviation, and delay deviation are multiplied by their respective weighting coefficients and summed to obtain a comprehensive deviation index. The confidence level of the test response signal is calculated based on the comprehensive deviation index. When the confidence level is lower than a preset confidence threshold, the sampling parameters are adjusted and the test response signal is re-acquired. The comprehensive deviation index is then output as the deviation value of the test response signal relative to the expected signal.

[0024] In one alternative implementation,

[0025] The steps of determining the fault test point based on the deviation value, analyzing the degree of impact on the test nodes connected to the fault test point based on the circuit connection relationship, and obtaining the test priority level of the test nodes include:

[0026] The deviation values ​​are constructed into a deviation feature vector. The deviation feature vector is standardized and weighted by the deviation weight coefficient to obtain a fault score. The fault test point is determined based on the fault score.

[0027] Calculate the electrical coupling coefficient and mutual inductance coupling coefficient between the fault test point and the adjacent test node. Multiply the electrical coupling coefficient and mutual inductance coupling coefficient by the coupling weight coefficient respectively and sum them to obtain the coupling coefficient matrix. Calculate the first-order influence strength of the fault test point on the adjacent test node based on the coupling coefficient matrix.

[0028] The path attenuation factor is calculated based on the connection distance between the test nodes. The product of the first-order influence intensity and the path attenuation factor is taken as the propagation influence intensity. Based on the propagation influence intensity, the multi-order influence intensity of the fault test point on non-adjacent test nodes is calculated. The influence intensity of different orders is multiplied by the corresponding influence weight coefficient and summed to obtain the total influence degree of the test node.

[0029] The basic priority is obtained by dividing the total impact by the maximum total impact. The timing adjustment coefficient is calculated based on the waiting time of the test node. The product of the basic priority and the timing adjustment coefficient is used as the test priority level of the test node.

[0030] Calculate the accuracy of the test priority level, and adjust the deviation weight coefficient, the coupling weight coefficient, and the influence weight coefficient according to the change in the accuracy.

[0031] In one alternative implementation,

[0032] The steps of calculating the path attenuation factor based on the connection distance between test nodes, using the product of the first-order influence intensity and the path attenuation factor as the propagation influence intensity, and calculating the multi-order influence intensity of the fault test point on non-adjacent test nodes based on the propagation influence intensity include:

[0033] Calculate the shortest physical connection path length between test nodes; obtain the frequency characteristics of the test response signal; calculate the frequency-related attenuation coefficient based on the frequency characteristics; use the product of the frequency-related attenuation coefficient and the shortest physical connection path length as the base attenuation value; calculate the branch loss value based on the number of branches in the propagation path; and combine the base attenuation value and the branch loss value to obtain the path attenuation factor.

[0034] Calculate the equivalent resistance value of the propagation path, determine the path propagation weight based on the equivalent resistance value, and take the product of the first-order influence intensity, the path attenuation factor, and the path propagation weight as the propagation influence intensity.

[0035] For each non-adjacent test node, the influence intensity propagated from the adjacent node is calculated based on the propagation influence intensity, and the influence intensity propagated from all adjacent nodes is superimposed to obtain the influence intensity of the current order.

[0036] In one alternative implementation,

[0037] The step of adjusting the test timing schedule according to the test priority level, and preferentially assigning the physical probes to high-priority test nodes, includes:

[0038] The occupancy status of probes within each time window is statistically analyzed, and the schedulable time period of physical probes is calculated based on the occupancy status. Test nodes are sorted according to their test priority level, and then assigned to schedulable time periods in descending order of priority. Based on the connection relationship between each controllable switch unit and the test node, the switching control sequence of the controllable switch unit is generated, and the test timing arrangement table is updated based on the switching control sequence.

[0039] Secondly, a multifunctional test point allocation system based on an adaptive switch matrix is ​​provided, including:

[0040] The first unit is used to acquire the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes circuit connection relationships;

[0041] The second unit is used to construct an adaptive switch matrix based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switch units in different time periods, thereby generating a test timing schedule.

[0042] The third unit is used to collect the test response signals of the logic test points, calculate the deviation value of the test response signals relative to the expected signals; determine the fault test points based on the deviation value, analyze the degree of impact on the test nodes connected to the fault test points based on the circuit connection relationship, and obtain the test priority level of the test nodes; adjust the test timing schedule table according to the test priority level, prioritize the allocation of the physical probes to the test nodes with higher priority levels, obtain the updated test timing schedule table, control the switching state of each controllable switch unit, and realize the dynamic allocation of test points.

[0043] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0044] This paper utilizes time-division multiplexing technology to achieve dynamic allocation of physical probes, which significantly improves the efficiency of test resource utilization.

[0045] A probe allocation strategy is constructed using machine learning and optimization algorithms, taking into account test point relevance and resource constraints, to maximize test coverage. An innovative approach incorporates fault propagation analysis, enabling the analysis of the impact of faulty test points on surrounding nodes and scientifically determining the priority of test nodes, thus concentrating limited probe resources on critical areas.

[0046] By employing multi-dimensional signal analysis techniques, comprehensively considering time-domain deviation, frequency-domain deviation, and delay deviation, the accuracy of fault detection is improved. The system can adjust test strategies in real time based on test responses, establishing an adaptive optimization mechanism to ensure the test process can be dynamically adjusted according to actual conditions. Through the introduction of coupling coefficient matrices and path attenuation factors, the system systematically evaluates fault propagation paths and their impact, achieving precise fault location. Furthermore, the system considers the waiting time of test nodes, introducing timing adjustment coefficients to prevent certain test points from being untested for extended periods, ensuring the fairness and comprehensiveness of the test. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the multifunctional test point allocation method based on an adaptive switch matrix according to an embodiment of the present invention.

[0048] Figure 2 This is a flowchart for determining fault test points and dynamically calculating test priorities based on deviation analysis. Detailed Implementation

[0049] The technical solutions of the present invention will be described below with reference to the accompanying drawings. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0050] Figure 1 This is a schematic diagram of the process of the method of the present invention, such as... Figure 1 As shown, the multifunctional test point allocation method based on the adaptive switch matrix includes:

[0051] Obtain the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes test point location information and circuit connection relationship;

[0052] An adaptive switch matrix is ​​constructed based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switch units in different time periods, thereby generating a test timing schedule.

[0053] The test response signals of the logic test points are collected, and the deviation value of the test response signals relative to the expected signals is calculated. Fault test points are determined based on the deviation value. The degree of influence on test nodes connected to the fault test points is analyzed based on the circuit connection relationship to obtain the test priority level of the test nodes. The test timing schedule is adjusted according to the test priority level, and the physical probes are preferentially allocated to test nodes with higher priority levels to obtain an updated test timing schedule. The switching state of each controllable switch unit is controlled to achieve dynamic allocation of test points.

[0054] For example, circuit topology information includes test point location information and circuit connection relationships. The test point location information describes the physical coordinates of each test point on the integrated circuit, such as the coordinates of a test point being (X=1250 micrometers, Y=850 micrometers). The circuit connection relationships describe the physical connections and logical relationships between the test points, including wire connections, functional module affiliation, and other information. Methods for obtaining circuit topology information include reading circuit design files (such as netlist files), layout files, and test planning files. For example, from the netlist file, it can be extracted that node 12 and node 18 are connected by a 5-ohm resistor, belonging to the same memory module; node 18 and node 25 are directly connected by a 180-micrometer-long wire, belonging to different functional modules. Circuit topology information should also include the electrical characteristics of each test point, such as the operating voltage range (e.g., 1.8V±0.1V), operating frequency (e.g., 100MHz), and signal type (e.g., differential signal).

[0055] An adaptive switch matrix is ​​constructed based on the acquired circuit topology information to establish a controllable connection network between physical probes and logic test points. The adaptive switch matrix consists of multiple controllable switch units, each of which can be considered an electronic switch capable of switching states according to control signals. These controllable switch units can be implemented using analog multiplexers, crossbar switch arrays, or relay matrices. For example, for a unit with 32 test points and 8 physical probes, a 4×8 switch matrix can be designed, allowing each physical probe to be connected to 4 different test points. The topology of the switch matrix needs to consider the spatial distribution of the test points; adjacent test points tend to be connected to the same group of switch units to reduce wiring complexity.

[0056] Efficient utilization of physical probe resources is achieved through time-division multiplexing. Time-division multiplexing refers to dynamically allocating the same physical probe to different logic test points within different time periods by controlling the switching states of controllable switch units in the switch matrix. For example, physical probe P1 is connected to test point N5 in time period T1, to test point N12 in time period T2, and to test point N17 in time period T3. The time period division should consider the signal stabilization time, the measurement time required, and the switch switching time. For high-frequency digital signals, shorter time periods (e.g., 1 microsecond) are needed; for slow analog signals, longer time periods (e.g., 5 microseconds) are needed. The generated test timing schedule table records the complete test plan, including the allocation of each probe, the test sequence, and the switch control signals within each time period.

[0057] The test response signals from the acquired logic test points are converted from analog to digital signals by an analog-to-digital converter for processing. The deviation between the test response signal and the expected signal reflects the operating status of the test point; excessive deviation indicates a fault. Based on the deviation, the faulty test point is identified, and the impact range of the fault is analyzed through circuit connections. The degree of impact on test nodes connected to the faulty test point is assessed, and the test priority of the test nodes is determined. For example, if node A is identified as a faulty test point, node B, which is directly connected to node A, is significantly affected and receives a higher test priority; while node C, which is farther from node A, is less affected and has a lower priority.

[0058] The test timing schedule is adjusted according to test priority levels, prioritizing the allocation of physical probes to higher-priority test nodes to ensure timely testing of critical nodes. The updated test timing schedule requires adjustments to the control timing of the switch matrix to implement the new probe allocation scheme. For example, if the priority level of node B increases from 0.3 to 0.8, in the updated timing schedule, node B will be tested earlier, moving from time period T8 to time period T2. This necessitates corresponding adjustments to the switching states of time periods T2 and T8. Based on the updated test timing schedule, a sequence of switch control signals is generated to drive each switch unit in the switch matrix to switch states according to a predetermined timing sequence. Switch control signals can be generated using a dedicated controller or FPGA to ensure timing accuracy. During switch switching, anti-jitter and settling time must be considered to avoid measurement errors caused by switching transients. For example, for a certain switch unit, a 50-nanosecond wait is required after switching to ensure signal stability; these timing parameters should be included in the calculation of the test timing schedule.

[0059] In one alternative implementation,

[0060] The steps for generating a test timing schedule by controlling the switching state of a controllable switching unit to allocate the same physical probe to different logic test points at different time periods include:

[0061] Obtain the feature vectors of the logical test points, which include signal frequency features, signal amplitude features, timing requirement features, and coupling coefficient features; calculate the correlation matrix between the test points based on the feature vectors of the logical test points; construct a probe allocation state space based on the correlation matrix, which includes probe allocation state, resource constraints, and test quality indicators; and construct a probe allocation action space, which includes probe number, test point number, and allocation time period.

[0062] A reward function is constructed based on the probe allocation state space and the probe allocation action space. The reward function includes a test coverage reward, a test quality reward, and a constraint violation penalty. The probe allocation strategy is iteratively optimized based on the reward function. The probe allocation strategy satisfies the single-moment probe allocation constraint, the test point correlation constraint, and the minimum test time constraint.

[0063] The system collects response signals from test points and calculates test quality assessment values. When the test quality assessment value is lower than a preset assessment threshold, it updates the weight parameters in the reward function based on the gradient descent method and re-executes the iterative optimization of the probe allocation strategy. A test timing schedule is generated based on the optimized probe allocation strategy. The test timing schedule includes time window division information, probe allocation scheme information, switch state control sequence information, and quality monitoring index information. The system controls the switch state based on the test timing schedule to achieve dynamic allocation of test points.

[0064] For example, the feature vector of a logic test point consists of signal frequency characteristics, signal amplitude characteristics, timing requirement characteristics, and coupling coefficient characteristics. The signal frequency characteristics reflect the main frequency components of the signal required by the test point; for example, a test point may require acquiring a signal with an operating frequency of 500MHz. The signal amplitude characteristics reflect the signal strength range of the test point; for example, the signal voltage range of a test point may be 0.8V to 1.2V. The timing requirement characteristics include the rise time, fall time, and hold time requirements of the test point signal; for example, the rise time must not exceed 2ns, the fall time must not exceed 3ns, and the hold time must be at least 15ns. The coupling coefficient characteristics reflect the degree of mutual interference between the test point and other test points; for example, the capacitive coupling coefficient between two adjacent test points is 0.015pF, and the inductive coupling coefficient is 0.023nH.

[0065] The correlation matrix between test points is calculated based on these feature vectors, including physical distance correlation, functional correlation, and temporal correlation. For physical distance correlation, the correlation coefficient for test points within 0.5 mm can be set to 0.85, for those between 0.5 mm and 1 mm it can be set to 0.65, for those between 1 mm and 2 mm it can be set to 0.45, and for those more than 2 mm it can be set to 0.25. Functional correlation is determined based on the functional module to which the test point belongs. The correlation coefficient for test points within the same functional unit can be set to 0.9, for those between adjacent functional units it can be set to 0.7, and for those without direct functional connection it can be set to 0.3. Temporal correlation is determined based on the signal propagation path. For example, the correlation coefficient for adjacent test points on the signal propagation path can be set to 0.95, for those separated by one node it can be set to 0.75, and for those separated by two nodes it can be set to 0.55.

[0066] The correlation matrix is ​​converted into test point grouping information. Test points with a correlation higher than 0.75 are grouped into the same test group to avoid simultaneous testing. For 32 logical test points, 12 test groups are formed, with the largest group containing 5 highly correlated test points and the smallest group containing only a single test point. The core of the probe allocation state space is the probe allocation state matrix, represented as an 8×32 matrix (assuming 8 physical probes and 32 logical test points), where an element value of 1 indicates a connection and 0 indicates no connection. The correlation matrix is ​​directly mapped to a set of constraints in the state space. For example, the test point pair (7,15) with a correlation value of 0.92 generates the constraint "Test point 7 and test point 15 cannot be tested in adjacent time windows". These constraints are encoded as state transition rules, restricting feasible change paths for probe allocation states. Resource constraints include probe quantity constraints (e.g., physical probes are limited to 8), time window constraints (total time windows are set to 64), and physical connection constraints of the switch matrix (some probes cannot connect to specific test points due to physical layout limitations). Test quality metrics are integrated into the state evaluation function, including test coverage (target 98% coverage of all test points), signal integrity (signal-to-noise ratio not less than 35dB), and test accuracy (error rate not exceeding 0.2%). The state space also includes a correlation-based test point priority mapping. The sum of the rows and columns of the correlation matrix is ​​used to calculate the criticality of test points; test points with high criticality (e.g., criticality value 90) receive higher test priority. The entire state space construction process establishes a direct mapping relationship from the correlation matrix to probe allocation strategy constraints, forming a complete state representation system guiding probe allocation decisions.

[0067] The probe allocation action space comprises three dimensions: probe number, test point number, and allocation time period. The probe number ranges from 1 to 8, corresponding to 8 physical probes. The test point number ranges from 1 to 32, corresponding to 32 logical test points. The allocation time period is in clock cycles, ranging from 1 to 64, representing 64 time windows. Each element in the action space represents an allocation decision, such as allocating probe 2 to test point 7 in time window 15.

[0068] In the reward function, the calculation formula for the test coverage reward is: Coverage Reward = 100 × (Number of Tested Points / Total Number of Test Points). For example, covering 30 out of 32 test points yields a score of 93.75. The test quality reward is composed of a weighted average of the signal-to-noise ratio (SNR) score and the error rate score. The SNR score is calculated as (Actual SNR / Target SNR) × 50, and the error rate score is calculated as (1 - Actual Error Rate / Maximum Allowable Error Rate) × 50. For example, an SNR of 40dB (target 35dB) yields 57.14 points, and an error rate of 0.15% (allowable 0.2%) yields 37.5 points. With weights of 0.6 and 0.4 respectively, the total quality reward is 49.29 points. The specific calculation method for the constraint violation penalty is as follows: Single-moment probe reassignment penalty = -20 × Number of violations; Correlation conflict penalty = -30 × Correlation coefficient × Number of violations; Time limit violation penalty = -1.5 × Exceedance percentage. If a probe is repeatedly assigned twice at a single moment, the penalty is -40; if test points with a correlation coefficient of 0.8 are tested simultaneously, the penalty is -24; if the test time exceeds 10%, the penalty is -15.

[0069] The iterative optimization process uses reinforcement learning to adjust the probe allocation scheme through multiple iterations. In each iteration, an allocation action is selected based on the current state, the reward value is calculated, and the policy parameters are updated. For example, under the initial policy, the average testing time for 32 test points is 45 time windows, the coverage is 85%, and the total reward value is 105. After 50 iterations, the average testing time is reduced to 32 time windows, the coverage increases to 97%, and the total reward value increases to 165. The optimization results satisfy the single-moment probe allocation constraint, the test point correlation constraint, and the minimum testing time constraint. The single-moment probe allocation constraint ensures that each probe connects to only one test point at a time. The test point correlation constraint ensures that test points with a correlation higher than 0.8 are not tested in adjacent time windows. The minimum testing time constraint ensures that the total time required to complete the testing of all necessary test points is minimized.

[0070] The response signal acquisition uses an analog-to-digital converter with a sampling rate of 2 GS / s and a resolution of 12 bits. Test quality assessment is based on three aspects: signal integrity, measurement accuracy, and coverage comprehensiveness. Signal integrity is evaluated using the signal-to-noise ratio (SNR). For example, an SNR of 32 dB, lower than the target of 35 dB, scores 85 points. Measurement accuracy is evaluated by deviation from the standard reference value. For example, an average deviation of 0.25%, higher than the target of 0.2%, scores 90 points. Coverage comprehensiveness is evaluated by the test point coverage rate. For example, a coverage rate of 94%, lower than the target of 98%, scores 92 points. The overall test quality assessment value is a weighted average of the three indicators, with weights of 0.35, 0.35, and 0.3, respectively, resulting in an overall assessment value of 88.85 points. A preset assessment threshold of 90 points is set. The current assessment value of 88.85 points is lower than the threshold, requiring an update of the weight parameters. The initial weights are set as follows: 0.4 for the test coverage reward item, 0.4 for the test quality reward item, and 0.2 for the constraint violation penalty item. Calculated using gradient descent, the weight adjustment increments were +0.05, +0.03, and -0.08, resulting in weights of 0.45, 0.43, and 0.12. The iterative optimization of the probe allocation strategy was then re-executed using the updated weights.

[0071] The probe-test-time point triplet data in the probe allocation strategy is structured to construct an initial allocation matrix. For 8 physical probes and 32 test points, each allocation decision in the optimization strategy is traversed, and the corresponding matrix element is set to 1. Then, time windows are divided, and the duration of each time window is determined based on the timing characteristics of the test points. For test points requiring stable signals, such as analog signal test points, a longer time window (e.g., 5 μs) is allocated; for digital logic test points, a shorter time window (e.g., 2 μs) can be allocated. The cumulative time is automatically calculated to determine the starting point of each window; for example, time window 1 starts at 0 μs, and window 2 starts at 5 μs. Probe allocation scheme information is directly obtained by querying the allocation matrix, recording the connection relationship between each probe and test point within each time window. Subsequently, a switch state control sequence is generated based on the probe allocation scheme and the physical connection topology of the switch matrix. This process requires querying the switch connection table, which stores the connection relationship between each switch unit and the probe and test point. For example, if probe 1 needs to connect to test point 5, the query reveals that switch units A1 and B3 need to be closed; therefore, these switches are set to closed in the corresponding time window. Simultaneously, quality monitoring metrics are calculated, including test coverage, signal integrity prediction, and timing margin. Potential conflicts are detected, such as interference caused by simultaneous testing of adjacent high-frequency test points; in such cases, the timing schedule is automatically adjusted or additional shielding measures are implemented. Finally, a binary control sequence containing complete timing information is generated to drive the switch matrix hardware to execute the testing process.

[0072] This invention significantly improves the utilization efficiency and test coverage of physical probes by constructing an adaptive probe allocation mechanism. The reinforcement learning-based optimization method can adapt to the testing requirements of different integrated circuits and dynamically adjust the probe allocation strategy. Through multi-dimensional feature vector analysis and correlation calculation, the correlation between test points is effectively identified, avoiding test interference. A closed-loop quality assessment and adaptive parameter adjustment mechanism is introduced to ensure continuous optimization of test quality.

[0073] In one alternative implementation,

[0074] The probe allocation strategy is iteratively optimized based on the reward function. The steps of the probe allocation strategy satisfying the single-time probe allocation constraint, the test point correlation constraint, and the minimum test time constraint include:

[0075] A binary decision variable is established for the probe allocation constraint at a single time step. The binary decision variable represents the allocation relationship between the probe and the test point at each time step. The degree of violation of the correlation constraint of the test point is calculated based on the binary decision variable. The degree of violation is obtained by multiplying the correlation matrix and the test state of the test point.

[0076] Normalize the degree of each constraint violation to obtain the normalized constraint violation degree. Dynamically update the constraint weights based on the normalized constraint violation degree. The update of the constraint weights is achieved by multiplying the learning rate and the normalized constraint violation degree. Calculate the comprehensive evaluation index of constraint satisfaction based on the constraint weights.

[0077] A neighborhood structure for probe allocation schemes is constructed, which includes a probe allocation time adjustment scheme and a probe and test point redistribution scheme. Historical schemes are recorded to form a taboo table, and the probe allocation scheme is locally optimized based on the taboo table.

[0078] The optimization convergence is calculated based on the comprehensive evaluation index, and the optimization step size is dynamically adjusted based on the optimization convergence. The comprehensive evaluation index is used as the optimization basis for the reward function.

[0079] Calculate the variance of the probe allocation scheme after multiple consecutive optimizations. When the variance is less than a preset variance threshold or the number of consecutive optimizations reaches a preset number of iterations, determine the optimization result of the probe allocation strategy.

[0080] For example, this specific embodiment discloses a method for iteratively optimizing a probe allocation strategy based on a reward function. This method ensures that the probe allocation strategy satisfies single-moment probe allocation constraints, test point correlation constraints, and minimum test time constraints. In practical implementation, this method is applicable to the field of integrated circuit testing, especially when physical probe resources are limited but logic test points are numerous.

[0081] In implementation, the binary decision variables of the single-moment probe allocation constraint represent the allocation relationship between probes and test points at each moment. These variables are in the form of a three-dimensional matrix, with dimensions corresponding to the probe number, test point number, and time period number, respectively. Taking a cell with 8 physical probes, 32 test points, and 64 time periods as an example, the decision variable matrix is ​​8×32×64. Each element in the matrix takes a value of 0 or 1. A value of 1 indicates that the corresponding probe is allocated to a specific test point within a specified time period, while a value of 0 indicates that it is not allocated. For example, if the decision variable at position (3,15,27) has a value of 1, it means that the 3rd probe is allocated to the 15th test point in the 27th time period. The single-moment probe allocation constraint requires that each probe can only be allocated to one test point within the same time period. This is manifested in the decision variable matrix where only one element in the plane corresponding to each probe and time period has a value of 1, while the rest are 0.

[0082] When calculating the degree of violation of the correlation constraints of test points based on binary decision variables, it is necessary to check whether test points with high correlation were tested in adjacent time periods. The testing status of each test point in each time period is extracted from the decision variables. If test point i is tested in time period t and test point j is tested in time period t+1, the correlation value between these two test points is queried from the correlation matrix and used as the degree of violation. The correlation matrix is ​​a two-dimensional array where each element represents the correlation coefficient between the corresponding two test points, with a value ranging from 0 to 1, determined by objective factors such as the physical distance and electrical connection of the test points. For example, assuming the correlation between test point 5 and test point 12 is 0.85, if the decision variables show that test point 5 was tested in time period 20 and test point 12 was tested in time period 21, then the degree of violation is 0.85. The overall degree of violation is obtained by summing the violations for all test point pairs and adjacent time periods.

[0083] The minimum test time constraint is calculated based on the sum of the test durations for each test point to ensure the total test time is minimized. Each test point has a minimum test duration requirement; for example, a fast signal test point requires 2 time slots, and a slow signal test point requires 5 time slots. The actual number of time slots allocated to each test point is calculated using decision variables. If it is less than the minimum requirement, a violation is constituted. For example, if a test point requires a minimum of 3 time slots but is only allocated 1 time slot, the violation level is 2 time slots. Simultaneously, the optimization objective is to reduce the total test time, that is, to minimize the total number of time slots required to complete the test for all test points.

[0084] The severity of each constraint violation is normalized by dividing it by its maximum severity. For example, the maximum violation severity for a single-moment probe allocation constraint is the number of probes × the number of time periods, i.e., 8 × 64 = 512; the maximum violation severity for a test point correlation constraint is the sum of correlations × the number of time periods; and the maximum violation severity for a minimum test time constraint is the sum of the minimum required times for all test points. After normalization, the severity of each constraint violation ranges from 0 to 1. For example, the normalized violation severity for a single-moment constraint is 0.05, for a correlation constraint it is 0.12, and for a minimum test time constraint it is 0.08.

[0085] Constraint weights determine the importance of each constraint in the optimization process. Initial constraint weights can be set to equal values, such as a single-time constraint weight of 0.33, a correlation constraint weight of 0.33, and a minimum-time constraint weight of 0.34. During optimization, the weights are dynamically adjusted based on the degree of violation; constraints with higher violation degrees have increased weights, while those with lower violation degrees have decreased weights. Weight updates are achieved by multiplying the learning rate by the normalized constraint violation degree. The learning rate controls the speed of weight adjustment; a smaller learning rate, such as 0.05, ensures the stability of the optimization process, while a larger learning rate, such as 0.2, accelerates convergence. For example, when the normalized violation degrees are 0.05, 0.12, and 0.08, using a learning rate of 0.1 for updates results in weight adjustments of 0.005, 0.012, and 0.008, with adjusted weights of 0.335, 0.342, and 0.323. A comprehensive evaluation index of constraint satisfaction is calculated based on the updated constraint weights by subtracting the weighted sum of each constraint violation degree and its corresponding weight from 1. For example, the comprehensive evaluation index = 1 - (0.05 × 0.335 + 0.12 × 0.342 + 0.08 × 0.323) = 0.918.

[0086] The neighborhood structure of the probe allocation scheme defines how to generate new candidate schemes from the current scheme. The probe allocation time adjustment scheme involves changing the allocation time of probes, such as changing probe 3, originally allocated to test point 8 in time period 10, to be allocated in time period 15. The probe and test point reassignment scheme involves changing the correspondence between probes and test points, such as changing probe 2, originally allocated to test point 5, to be allocated to test point 9. When generating neighborhood schemes, the priority and relevance of test points are considered. High-priority test points, such as those in the processor core region, have a higher allocation probability; highly relevant test points are avoided in adjacent time periods. The neighborhood size is related to the problem size; for example, for 8 probes, 32 test points, and 64 time periods, generating 10-20 neighborhood schemes each time is appropriate. The tabu list records the characteristics of recently visited schemes to prevent the search process from getting stuck in a loop. The tabu list size can be set to 15-20, recording the characteristics of the most recent 15-20 schemes. Scheme characteristics can be simplified into combinations of key allocation decisions; for example, the decision to allocate probe 3 to test point 6 in time period 12 can be represented as (3, 6, 12). When a new neighborhood solution is generated, its features are checked against the tabu list. If they are, the solution is skipped unless its evaluation metric is significantly better than the current best solution. Local optimization employs an iterative improvement strategy based on tabu search. In each iteration, the solution with the best evaluation metric among the neighborhood solutions that is not in the tabu list is selected as the new current solution. The tabu list is then updated, and the iteration continues. For example, if the current solution has an overall evaluation metric of 0.918, the best neighborhood solution has an evaluation metric of 0.935, and this solution is not in the tabu list, then this solution is accepted, and the current solution is updated.

[0087] The optimization convergence is the difference between the current evaluation metric and the previous evaluation metric. For example, if the current evaluation metric is 0.935 and the previous one was 0.918, the improvement is 0.017. The optimization step size controls the magnitude of change in the neighborhood structure. A larger step size allows for larger changes in the solution, which is beneficial for escaping local optima; a smaller step size is suitable for fine-tuning and improving convergence accuracy. When the improvement is significant, the step size can be reduced for fine-tuning; when the improvement is small or negative, the step size can be increased to escape local optima. For example, when the improvement is 0.017, the step size can be reduced from the initial value of 0.2 to 0.15; when the improvement is less than 0.005 for 5 consecutive rounds, the step size can be increased to 0.25. The comprehensive evaluation metric serves as the basis for optimizing the reward function, and the probe allocation strategy is adjusted after each iteration based on the changes in the evaluation metric.

[0088] Calculate the variance of the probe allocation scheme after multiple consecutive optimizations to determine if the optimization has converged. The scheme variance reflects the stability of the optimization results over multiple iterations. Variance calculation is based on key performance indicators, such as the comprehensive evaluation index and the degree of constraint violation. For example, record the comprehensive evaluation index values ​​for the most recent 10 iterations: 0.918, 0.935, 0.942, 0.946, 0.951, 0.953, 0.955, 0.955, 0.956, 0.956, and calculate the variance of these values ​​as 0.00015. Set a preset variance threshold of 0.0001. When the variance is less than this threshold or the number of consecutive optimizations reaches a preset number of iterations (e.g., 100), the optimization is considered to have converged, iteration stops, and the current probe allocation strategy is output as the final optimization result. The final optimization result includes the allocation of each probe in each time period, as well as the corresponding constraint satisfaction and comprehensive evaluation index. For example, the final solution has a single-moment constraint violation rate of 0.01, a correlation constraint violation rate of 0.03, a minimum time constraint violation rate of 0.02, and a comprehensive evaluation index of 0.956.

[0089] This invention effectively solves the complex constraint satisfaction problem of probe resource allocation by establishing binary decision variables and a dynamic weight adjustment mechanism. The tabu search-based local optimization strategy avoids the search process getting trapped in local optima, thus improving the quality of the solution.

[0090] In one alternative implementation,

[0091] The steps for calculating the deviation of the test response signal from the expected signal include:

[0092] The test response signal is filtered to obtain a filtered signal. The filtered signal is divided into signal segments according to a preset time window. An expected signal template is established based on a standard waveform. The root mean square error between the signal segment and the expected signal template is calculated to obtain the time domain deviation. The spectrum between the signal segment and the expected signal template is calculated to obtain the frequency domain deviation. The delay deviation is calculated based on the cross-correlation function between the signal segment and the expected signal template.

[0093] The time-domain deviation, frequency-domain deviation, and delay deviation are multiplied by their respective weighting coefficients and summed to obtain a comprehensive deviation index. The confidence level of the test response signal is calculated based on the comprehensive deviation index. When the confidence level is lower than a preset confidence threshold, the sampling parameters are adjusted and the test response signal is re-acquired until the confidence level is greater than or equal to the preset confidence threshold. The comprehensive deviation index is then output as the deviation value of the test response signal relative to the expected signal.

[0094] For example, a bandpass filter is used for the test response signal, and the filter parameters are selected according to the signal characteristics. For instance, for integrated circuit test signals with a frequency range of 1MHz to 50MHz, the passband of the bandpass filter can be set to 0.9MHz to 55MHz, and the stopband attenuation to 60dB.

[0095] Dividing the filtered signal into segments according to a preset time window facilitates segmented analysis and evaluation. The size of the time window depends on the characteristics of the signal being measured; generally, the smallest window that can encompass the complete cycle or characteristics of the signal is chosen. For example, for clock signals, a window containing 10-20 complete cycles can be selected; for burst signals, a window containing the complete burst process can be selected. In practical applications, if the signal period is 10ns, the time window can be set to 200ns, i.e., containing 20 complete cycles. Signal segments can overlap to some extent to enhance the continuity of the analysis. For example, adjacent signal segments can overlap by 25% of their length; that is, for a 200ns time window, the starting points of adjacent windows differ by 150ns.

[0096] The standard waveform can be a design specification waveform, a theoretical analysis waveform, or a validated, high-quality reference waveform. The expected signal template should include the signal's key characteristics, such as amplitude, frequency, phase, rise / fall time, and duty cycle. The template construction method differs depending on the type of signal. For example, for clock signals, the template could be an ideal square wave; for analog signals, the template could be a sine wave, a triangle wave, etc. The expected signal template should also consider a reasonable range of variation under actual test conditions to accommodate inherent test errors. For example, the expected signal template may allow for 5% amplitude variation and 1% phase shift.

[0097] The signal segment and the expected signal template are time-aligned to ensure they match at the same starting point. Then, the squared difference between the actual and expected values ​​is calculated for each sampling point. All squared differences are summed and divided by the number of sampling points to obtain the average value. Finally, the square root of this average value is taken to obtain the root mean square error (RMSE), which serves as the time-domain deviation indicator. A smaller time-domain deviation indicates that the test response signal is closer to the expected signal in waveform shape. For example, if a signal segment contains 1000 sampling points and the calculated RMS error is 0.05V, it means that the average deviation of this signal segment from the expected signal template in the time domain is 0.05V.

[0098] Spectral analysis is performed on both the signal segment and the expected signal template to obtain their respective frequency components. Then, the differences between the two at various frequency points are compared, with a focus on the amplitude differences of the signal's dominant frequency and its harmonic components. Frequency domain deviation can be expressed as a weighted sum of the amplitude differences at each frequency point, with the weights set according to the importance of the frequency. For example, higher weights can be set for the dominant frequency and its lower-order harmonics, while lower weights can be set for high-frequency noise components. Frequency domain deviation can effectively identify problems such as frequency distortion, harmonic distortion, and noise interference in the signal. For example, if the amplitude of the test response signal at the dominant frequency of 10MHz is 5% lower than expected, and the second harmonic at 20MHz is 8% higher than expected, it indicates that the signal has nonlinear distortion.

[0099] The cross-correlation function describes the relationship between the similarity between two signals and the time offset. The correlation coefficient is calculated for the signal segment and the expected signal template at different time offsets. The time offset that maximizes the correlation coefficient is the delay bias. The delay bias reflects the time offset of the test response signal relative to the expected signal, and is particularly important for timing-sensitive circuit testing. For example, if the calculated maximum correlation coefficient corresponds to a time offset of 2ns, it indicates that the test response signal has a 2ns delay relative to the expected signal. Besides the delay value, the magnitude of the maximum correlation coefficient also reflects the similarity between the two signals and can be used as an auxiliary indicator for signal quality assessment.

[0100] The overall deviation index is obtained by multiplying the time-domain deviation, frequency-domain deviation, and delay deviation by their respective weighting coefficients and summing the results. The weighting coefficients are set according to the sensitivity of different application scenarios to each type of deviation. For example, for timing-critical circuits, a higher weight can be assigned to delay deviation, such as 0.5; for analog circuits with high waveform quality requirements, a higher weight can be assigned to time-domain deviation, such as 0.4; and for circuits sensitive to specific frequency responses, a higher weight can be assigned to frequency-domain deviation. The sum of the weighting coefficients is usually set to 1 to maintain the consistency of the numerical scale of the overall deviation index. For example, if the weight of time-domain deviation is set to 0.4, frequency-domain deviation to 0.3, and delay deviation to 0.3, and the calculated time-domain deviation is 0.05, frequency-domain deviation to 0.08, and delay deviation to 0.03, then the overall deviation index is 0.05 × 0.4 + 0.08 × 0.3 + 0.03 × 0.3 = 0.053.

[0101] Confidence level is the inverse function of the composite deviation index; the smaller the deviation, the higher the confidence level. It can be calculated by mapping the composite deviation index to a confidence level range between 0 and 1. For example, confidence level = 1 - composite deviation index / maximum permissible deviation. The maximum permissible deviation is a threshold set according to the test requirements, representing the maximum acceptable level of deviation. For instance, if the maximum permissible deviation is set to 0.2 and the composite deviation index is 0.053, then the confidence level = 1 - 0.053 / 0.2 = 0.735, indicating that the test result has a 73.5% confidence level.

[0102] When the confidence level falls below the preset confidence threshold, the sampling parameters need to be adjusted and the test response signal reacquired. The preset confidence threshold is determined based on the test quality requirements; for example, it can be set to 0.8, indicating that the test result must have at least 80% confidence. Adjusting the sampling parameters involves several aspects: increasing the sampling rate to improve the signal's temporal resolution; adjusting the probe position to improve signal acquisition quality; increasing the sampling duration to acquire more signal periods and improve statistical reliability; and adjusting filtering parameters to optimize noise suppression. For example, if the initial sampling rate is 100 MSa / s and the confidence level is 0.735, which is lower than the preset threshold of 0.8, the sampling rate can be increased to 200 MSa / s to reacquire the signal. After adjusting the parameters, the signal is reacquired, and the confidence level is recalculated until it reaches or exceeds the preset threshold. Once the confidence level meets the requirements, a comprehensive deviation index is output as the deviation value of the test response signal relative to the expected signal. This deviation value can be used for subsequent chip performance evaluation and product selection.

[0103] This invention employs an adaptive sampling strategy, dynamically adjusting test parameters based on confidence levels to ensure the reliability of test results.

[0104] In one alternative implementation,

[0105] The steps of determining the fault test point based on the deviation value, analyzing the degree of impact on the test nodes connected to the fault test point based on the circuit connection relationship, and obtaining the test priority level of the test nodes include:

[0106] The deviation values ​​are constructed into a deviation feature vector containing time domain, frequency domain, delay and amplitude. The deviation feature vector is standardized and weighted by the deviation weight coefficient to obtain the fault score. The fault test point is determined based on the fault score.

[0107] Calculate the electrical coupling coefficient and mutual inductance coupling coefficient between the fault test point and the adjacent test node. Multiply the electrical coupling coefficient and mutual inductance coupling coefficient by the coupling weight coefficient respectively and sum them to obtain the coupling coefficient matrix. Calculate the first-order influence strength of the fault test point on the adjacent test node based on the coupling coefficient matrix.

[0108] The path attenuation factor is calculated based on the connection distance between the test nodes. The product of the first-order influence intensity and the path attenuation factor is taken as the propagation influence intensity. Based on the propagation influence intensity, the multi-order influence intensity of the fault test point on non-adjacent test nodes is calculated. The influence intensity of different orders is multiplied by the corresponding influence weight coefficient and summed to obtain the total influence degree of the test node.

[0109] The basic priority is obtained by dividing the total impact by the maximum total impact. The timing adjustment coefficient is calculated based on the waiting time of the test node. The product of the basic priority and the timing adjustment coefficient is used as the test priority level of the test node.

[0110] Calculate the accuracy of the test priority level, and adjust the deviation weight coefficient, the coupling weight coefficient, and the influence weight coefficient according to the change in the accuracy until the accuracy is greater than a preset accuracy threshold.

[0111] For example, in combination Figure 2 The flowchart illustrating the fault test point determination and dynamic calculation of test priority based on deviation analysis is as follows: The deviation feature vector contains four dimensions: time-domain deviation reflects waveform shape differences, frequency-domain deviation reflects frequency component differences, delay deviation reflects timing synchronization differences, and amplitude deviation reflects signal strength differences. For example, the deviation feature vector of a certain test point is [0.05, 0.08, 2.5ns, 0.12], representing a root mean square deviation of 0.05 in the time domain, a spectral deviation of 0.08 in the frequency domain, a delay deviation of 2.5 nanoseconds, and an amplitude deviation of 0.12, respectively. The standardization method divides the deviation value of each dimension by the maximum permissible deviation value of that dimension, so that the standardized values ​​of each dimension are uniformly between 0 and 1. For example, if the maximum permissible time-domain deviation is 0.2, the frequency-domain deviation is 0.3, the delay deviation is 10ns, and the amplitude deviation is 0.5, then the standardized deviation feature vector is [0.25, 0.27, 0.25, 0.24]. The standardized feature vector is weighted by a preset deviation weight coefficient to obtain the fault score. The deviation weighting coefficient reflects the importance of different types of deviations in fault diagnosis. For example, the deviation weighting coefficient can be set to [0.3, 0.25, 0.25, 0.2], then the fault score is 0.25×0.3+0.27×0.25+0.25×0.25+0.24×0.2=0.253. Test points whose fault scores exceed a preset threshold (e.g., 0.2) are identified as fault test points.

[0112] For the connection between a fault test point and an adjacent test node, the conductivity coupling coefficient reflects the direct electrical connection relationship and can be calculated based on the resistance of the shared conductor and the connection method. For example, if test point A and test point B are connected by a conductor with a resistance of 10 ohms, and the reference resistance of the circuit is 100 ohms, then the conductivity coupling coefficient between them can be calculated as 0.1. The mutual inductance coupling coefficient reflects the electromagnetic induction relationship. The calculation method is as follows: the distance between parallel conductors is d, the parallel length is l, and the conductor diameter is r. The mutual inductance is M = μ0l / (2π) × [ln(2l / d) - 0.75], where μ0 is the free permeability 4π × 10⁻⁶. -7 H / m. For a spacing of 0.1mm, a parallel length of 5mm, and a wire diameter of 0.05mm, substituting these values ​​into the calculation yields M=1.15nH. Dividing by the normalization factor gives a mutual inductance coefficient of 0.05. The conductance coupling coefficient and the mutual inductance coupling coefficient are multiplied by the coupling weight coefficients and summed to obtain the comprehensive coupling coefficient matrix. The coupling weight coefficients depend on the circuit operating frequency and signal characteristics. For example, in high-frequency circuits, the weight of mutual inductance coupling is higher, such as a conductance weight of 0.6 and a mutual inductance weight of 0.4; in low-frequency circuits, the weight of conductance coupling is higher, such as a conductance weight of 0.8 and a mutual inductance weight of 0.2. The coupling coefficient matrix is ​​an N×N matrix (N is the number of test nodes), where each element represents the comprehensive coupling strength between two corresponding test nodes. Based on the coupling coefficient matrix, the first-order influence strength of the fault test point on adjacent test nodes is calculated, i.e., the degree of influence of the test nodes directly affected by the fault test point. For example, if the coupling coefficient between the fault test point F and the test node G is 0.15, then the first-order influence strength of F on G is 0.15.

[0113] The connection distance between test nodes can be either topological distance (number of nodes traversed) or physical distance (actual cabling length). The path attenuation factor typically decreases exponentially with distance; for example, an attenuation factor of 0.7 raised to the power of distance can be set. If the distance between two nodes is 2, the attenuation factor is 0.7 squared, or 0.49. The product of the first-order influence strength and the path attenuation factor is used to calculate the propagation influence strength, which is then used to calculate the multi-order influence strength of the fault test point on non-adjacent test nodes. The multi-order influence strength represents the degree to which the fault propagates to the distant node through multiple intermediate nodes. For example, if the first-order influence strength of fault test point F on node G is 0.15, the first-order influence strength of G on node H is 0.2, and the path attenuation factor between F and H is 0.49, then the second-order influence strength of F on H can be calculated as 0.15 × 0.2 × 0.49 = 0.0147. Similarly, the third-order, fourth-order, and higher-order influence strengths can be calculated. The total influence of the test node is obtained by multiplying the influence strengths of different orders by their corresponding influence weight coefficients and summing the results. The influence weight coefficients reflect the importance of different orders of influence. Generally, higher-order influences have lower weights. For example, the weight of a first-order influence can be set to 0.6, the weight of a second-order influence to 0.3, and the weight of a third-order influence to 0.1. If a test node is affected by a first-order influence of 0.15, a second-order influence of 0.03, and a third-order influence of 0.01, then the total influence is 0.15×0.6+0.03×0.3+0.01×0.1=0.1×0.6+0.03×0.3+0.01×0.1=0.099.

[0114] The base priority is obtained by dividing the total impact by the maximum total impact, ensuring that the priority of each node is between 0 and 1 for easy comparison and sorting. The maximum total impact is the maximum value of the total impact of all nodes in the current testing cycle, calculated by comparing all nodes, or it can be set as the theoretical maximum impact value. For example, if the maximum total impact is 0.25 and the total impact of a node is 0.099, then its base priority is 0.099 / 0.25 = 0.396. A timing adjustment coefficient is calculated based on the waiting time of the test nodes to prevent low-priority nodes from being untested for extended periods. Waiting time refers to the number of time intervals elapsed since the last test. The timing adjustment coefficient is designed as an increasing function of the waiting time, for example, represented as 1 plus the product of the waiting time and the time factor. If the time factor is set to 0.05, and a node has waited for 10 time intervals, then its timing adjustment coefficient is 1 + 10 × 0.05 = 1.5. The product of the base priority and the timing adjustment coefficient is used as the test priority level of the test node, considering both fault impact and timing fairness. For example, if the base priority is 0.396 and the timing adjustment factor is 1.5, then the test priority level is 0.396 × 1.5 = 0.594. Nodes with higher test priority levels will be scheduled for testing first.

[0115] The accuracy of priority ranking is determined by comparing the consistency of the current priority ranking with the actual fault distribution. For example, using historical fault data or expert-annotated real priorities as a reference, the consistency ratio between the current priority ranking and the reference ranking is calculated. If 16 out of the top 20 high-priority nodes are consistent with the reference ranking, the accuracy is 80%. The bias weight coefficient, coupling weight coefficient, and influence weight coefficient are adjusted based on the change in accuracy to optimize the priority allocation effect. The adjustment method uses gradient descent to gradually find the parameter combination that maximizes accuracy. For example, if the current accuracy is 80%, and adjusting the bias weight coefficient from [0.3, 0.25, 0.25, 0.2] to [0.35, 0.25, 0.2, 0.2] improves the accuracy to 82%, then the new weight coefficient is adopted. The parameters are continuously adjusted until the accuracy exceeds a preset accuracy threshold (e.g., 85%) or the maximum number of iterations is reached.

[0116] For example, taking a 32-node integrated circuit as an example, during the testing process, the deviation feature vector of node 7 was found to be [0.08, 0.12, 3.5ns, 0.15], which, after standardization, became [0.4, 0.4, 0.35, 0.3]. The weighted fault score was 0.367, exceeding the threshold of 0.2, and it was identified as a fault test point. The coupling coefficients of node 7 with adjacent nodes were calculated, yielding a conductance coupling of 0.22 and a mutual inductance coupling of 0.05 with node 9, resulting in a combined coupling coefficient of 0.22 × 0.7 + 0.05 × 0.3 = 0.169. Similarly, the coupling coefficients with other adjacent nodes were calculated to form a coupling coefficient matrix. Based on the coupling coefficients, the first-order influence strength of node 7 on adjacent nodes was calculated; for example, the first-order influence on node 9 was 0.169. Next, calculate the multi-order influence of non-adjacent nodes. For example, the coupling coefficient between node 9 and node 15 is 0.14, and the path attenuation factor from node 7 to node 15 is 0.7 squared = 0.49. Therefore, the second-order influence of node 7 on node 15 is 0.169 × 0.14 × 0.49 = 0.0116. Combine the influence of each order to obtain the total influence of each node. For example, the total influence of node 9 is 0.169 × 0.6 + 0.03 × 0.3 + 0.005 × 0.1 = 0.1143. Assuming the maximum total influence is 0.2, the basic priority of node 9 is 0.1143 / 0.2 = 0.5715. If node 9 has waited for 8 time periods, the timing adjustment coefficient is 1 + 8 × 0.05 = 1.4, then its final test priority is 0.5715 × 1.4 = 0.8001. Calculate the priority of all nodes using this method and sort them for testing.

[0117] This invention establishes a fault scoring mechanism based on multidimensional deviation characteristics and an influence propagation mechanism considering circuit topology, enabling accurate identification of fault test points and quantification of their impact on surrounding nodes. By introducing a timing adjustment mechanism, it balances test efficiency and fairness, preventing low-priority nodes from being ignored for extended periods. The adaptive parameter adjustment strategy ensures the accuracy and adaptability of priority allocation, improves test coverage and fault detection rate, reduces test resource waste, and provides an efficient and reliable node priority ranking method for integrated circuit testing.

[0118] In one alternative implementation,

[0119] The steps of calculating the path attenuation factor based on the connection distance between test nodes, using the product of the first-order influence intensity and the path attenuation factor as the propagation influence intensity, and calculating the multi-order influence intensity of the fault test point on non-adjacent test nodes based on the propagation influence intensity include:

[0120] Calculate the shortest physical connection path length between test nodes; obtain the frequency characteristics of the test response signal; calculate the frequency-related attenuation coefficient based on the frequency characteristics; use the product of the frequency-related attenuation coefficient and the shortest physical connection path length as the base attenuation value; calculate the branch loss value based on the number of branches in the propagation path; and combine the base attenuation value and the branch loss value to obtain the path attenuation factor.

[0121] Calculate the equivalent resistance value of the propagation path, determine the path propagation weight based on the equivalent resistance value, take the product of the first-order influence intensity, the path attenuation factor, and the path propagation weight as the propagation influence intensity, and determine the effective path affecting propagation based on the propagation influence intensity.

[0122] For each non-adjacent test node, the influence intensity propagated from the adjacent node is calculated based on the propagation influence intensity. The influence intensities propagated from all adjacent nodes are superimposed to obtain the influence intensity of the current order. The calculation stops when the influence intensity of the current order is less than one-tenth of the influence intensity of the previous order, thus obtaining the multi-order influence intensity.

[0123] For example, the physical connection path length between test nodes can be obtained from circuit layout information and expressed as the actual wire length, in micrometers or nanometers. In actual implementation, a connection graph can be constructed based on the circuit netlist and layout data, and the shortest path algorithm can be used to calculate the shortest physical distance between any two test nodes. For example, in a certain integrated circuit, the shortest physical connection path length between node A and node B is 250 micrometers, and the shortest physical connection path length between node B and node C is 180 micrometers. Then, the path length between node A and node C is 430 micrometers. If there are other paths, such as node A connecting to node C through node D, with a path length of 390 micrometers, then the shortest physical connection path length between node A and node C is 390 micrometers. For complex circuits, breadth-first search or Dijkstra's algorithm can be used to calculate the shortest path to ensure the accuracy of the path length calculation.

[0124] Frequency characteristics are obtained through spectral analysis of the test response signal, including the dominant frequency, bandwidth, and harmonic components. Signals of different frequencies exhibit different attenuation characteristics during propagation, with higher-frequency signals typically attenuating faster. The frequency-dependent attenuation coefficient can be expressed as an increasing function of frequency; for example, the attenuation coefficient can be set as the dominant frequency value divided by the reference frequency value and then multiplied by the attenuation base. Assuming the dominant frequency of the signal is 100MHz, the reference frequency is 10MHz, and the attenuation base is 0.05, then the frequency-dependent attenuation coefficient is 100 / 10 × 0.05 = 0.5. For complex signals containing multiple frequency components, a weighted average attenuation coefficient can be calculated, with the weights related to the energy proportion of each frequency component. For example, if the signal contains 60% of the 50MHz component and 40% of the 150MHz component, the reference frequency is 10MHz, and the attenuation base is 0.05, then the weighted average attenuation coefficient is (50 / 10×0.05×0.6)+(150 / 10×0.05×0.4)=0.15+0.3=0.45.

[0125] The base attenuation value is the product of the frequency-dependent attenuation coefficient and the shortest physical path length. This base attenuation value can be understood as the total attenuation of the signal propagating in a straight line. For example, if the frequency-dependent attenuation coefficient is 0.45 and the shortest physical path length is 390 micrometers, then the base attenuation value is 0.45 × 390 / 1000 = 0.1755. A larger base attenuation value indicates more severe signal attenuation. In practical circuits, signals not only propagate in straight lines but also shunt at branch points, resulting in additional attenuation. The branch loss value is calculated based on the number of branches in the propagation path. This branch loss value can be expressed as the product of the number of branches and the branch loss coefficient. The branch loss coefficient depends on the branch characteristics and is typically between 0.1 and 0.3. For example, if there are two branch points in the path and the branch loss coefficient is 0.2, then the branch loss value is 2 × 0.2 = 0.4. The path attenuation factor is obtained by combining the base attenuation value and the branch loss value. This combination can be achieved by multiplying the base attenuation value by 1 plus the branch loss value, or by adding the two together and taking the negative exponent. Using the former method, if the base attenuation value is 0.1755 and the branch loss value is 0.4, then the path attenuation factor is 0.1755×(1+0.4)=0.2457.

[0126] The equivalent resistance of the propagation path is calculated using circuit topology and component parameters, representing the total resistance the signal traverses from the source node to the target node. In a simplified model, it can be assumed that each unit length of wire has a fixed resistance, such as 0.1 ohms / 100 micrometers. Therefore, the equivalent resistance of a 390-micrometer path is 0.39 ohms. Considering the resistors, capacitors, and other components along the path, the actual equivalent resistance is larger. For example, if the path also includes a 10-ohm resistor and a component with an equivalent resistance of 5 ohms, the total equivalent resistance is 0.39 + 10 + 5 = 15.39 ohms. The propagation weight of the path is determined based on the equivalent resistance value. The propagation weight is inversely proportional to the equivalent resistance and can be expressed as the reference resistance value divided by the equivalent resistance value. If the reference resistance is 20 ohms and the equivalent resistance is 15.39 ohms, the path propagation weight is 20 / 15.39 = 1.3. The smaller the equivalent resistance value, the larger the propagation weight, indicating a greater contribution of the path to signal propagation.

[0127] The propagation influence strength is calculated by multiplying the first-order influence strength by the path attenuation factor and the path propagation weight. For example, if the first-order influence strength of a fault test point on its adjacent nodes is 0.8, the path attenuation factor is 0.2457, and the path propagation weight is 1.3, then the propagation influence strength is 0.8 × 0.2457 × 1.3 = 0.2556. Paths with a propagation influence strength greater than a preset threshold (e.g., 0.1) are identified as effective paths for influence propagation. Effective paths refer to paths where the fault influence can propagate significantly. Screening effective paths can reduce computational load and improve analysis efficiency.

[0128] For each non-adjacent test node, the influence intensity propagating from adjacent nodes is calculated based on the propagation influence intensity. The influence intensities propagated from all adjacent nodes are then summed to obtain the current order influence intensity. This step realizes the layer-by-layer propagation calculation of multi-order influences. For a non-adjacent node D, all nodes directly connected to D and adjacent to the fault test point F are considered, such as nodes B and C. The influence intensity of F propagating from B to D is calculated, which is the product of the first-order influence intensity of F on B and the propagation influence intensity from B to D. Similarly, the influence intensity of F propagating from C to D is calculated. These influence intensities are then summed to obtain the second-order influence intensity of F on D. For example, if the first-order influence intensity of F on B is 0.8, the propagation influence intensity from B to D is 0.3, the first-order influence intensity of F on C is 0.6, and the propagation influence intensity from C to D is 0.4, then the second-order influence intensity of F on D is 0.8 × 0.3 + 0.6 × 0.4 = 0.24 + 0.24 = 0.48. The calculation stops when the influence strength of the current order is less than one-tenth of the influence strength of the previous order, thus obtaining the multi-order influence strength. This termination condition is based on the characteristic that influence strength decays rapidly with increasing propagation order, avoiding unnecessary calculations. The multi-order influence strength is the sum of the influence strengths of each order, including direct influence (first order) and indirect influence (second order and above). For example, continuing the above case, let's calculate the third-order influence strength of F on E. If the propagation influence strength from D to E is 0.5, then the third-order influence strength of F propagating from D to E is 0.48 × 0.5 = 0.24. If the second-order influence strength of F on E is 2.5, then the third-order influence strength of 0.24 is less than one-tenth of the second-order influence strength (2.5 / 10 = 0.25), so the calculation of higher-order influences stops. The multi-order influence strength of F on E is the sum of the influence strengths of each order. If the first order is 0, the second order is 2.5, and the third order is 0.24, then the multi-order influence strength is 0 + 2.5 + 0.24 = 2.74.

[0129] Taking an integrated circuit with 64 test nodes as an example, assuming node 18 is identified as a fault test point, it is necessary to calculate the multi-order influence strength of its impact on other nodes. Node 18 is directly adjacent to nodes 23, 31, and 42, with first-order influence strengths of 0.75, 0.62, and 0.58, respectively. When calculating the impact of node 18 on the non-adjacent node 27, it is necessary to analyze the propagation path from node 18 to node 27. Nodes 23 and 42 are both directly connected to node 27, forming two paths from node 18 to node 27. The physical connection length from node 18 to node 23 is 120 micrometers, and the physical connection length from node 23 to node 27 is 150 micrometers, forming a path with a total length of 270 micrometers; the physical connection length from node 18 to node 42 is 180 micrometers, and the physical connection length from node 42 to node 27 is 80 micrometers, forming a path with a total length of 260 micrometers. Path 18-42-27 is slightly shorter and is the shortest physical connection path. If the main frequency of the test response signal is 80MHz, and the calculated frequency-dependent attenuation coefficient is 0.4, then the basic attenuation value is 0.4×260 / 1000=0.104. There is one branch point on the path, with a branch loss coefficient of 0.2, resulting in a branch loss value of 0.2×1=0.2. The path attenuation factor is 0.104×(1+0.2)=0.1248. The calculated equivalent resistance is 12 ohms, the reference resistance is 15 ohms, and the path propagation weight is 15 / 12=1.25. The first-order influence strength of node 18 on node 42 is 0.58, so the propagation influence strength through path 18-42-27 is 0.58×0.1248×1.25=0.09048. Similarly, the propagation influence strength of path 18-23-27 is calculated as 0.75×0.135×1.2=0.1215. Therefore, the second-order influence strength of node 18 on node 27 is 0.09048 + 0.1215 = 0.21198. We continue calculating higher-order influences until the termination condition is met, thus obtaining the multi-order influence strengths of node 18 on node 27.

[0130] The multi-order impact calculation method of this invention effectively captures the cascading propagation effect of faults in complex circuits, while the intelligent termination condition optimizes the calculation efficiency, realizes the accurate quantification of fault impact propagation, provides a reliable basis for prioritizing test nodes, significantly improves the pertinence and efficiency of integrated circuit testing, reduces test blind spots, and enhances fault detection capabilities and circuit reliability assessment levels.

[0131] In one alternative implementation,

[0132] The step of adjusting the test timing schedule according to the test priority level, and preferentially assigning the physical probes to high-priority test nodes, includes:

[0133] The occupancy status of probes within each time window is statistically analyzed, and the schedulable time periods for physical probes are calculated based on the occupancy status. Test nodes are sorted according to their test priority level, and then assigned to schedulable time periods in descending order of priority. When time period conflicts occur, lower-priority test nodes are postponed to subsequent available time periods. Based on the connection relationship between each controllable switch unit and the test node, the switching control timing of the controllable switch unit is generated, and the test timing arrangement table is updated based on the switching control timing.

[0134] For example, the occupancy status of probes within each time window is statistically analyzed. A time window refers to a basic time unit in the testing process, typically measured in microseconds or nanoseconds. The probe occupancy status indicates the usage of each physical probe within each time window and can be represented by binary flags: 1 indicates occupied, and 0 indicates available. For instance, for a test unit with 8 physical probes, an 8×T occupancy status matrix can be constructed, where T is the total number of time windows. Assume there are 8 probes P1 to P8, and the test is divided into 10 time windows T1 to T10. Initially, P1 is occupied in time windows T1 to T3, P2 is occupied in time windows T2 to T4, and the remaining probes and time windows are available. Based on the occupancy status, the schedulable time periods for each physical probe are calculated, i.e., the consecutive available time periods for each probe. For example, the schedulable time period for probe P1 is [T4-T10], the schedulable time periods for probe P2 are [T1-T1] and [T5-T10], and so on. At the same time, the required test duration for each test node is also considered. For example, if a test node needs to complete the test in 3 consecutive time windows, then only probes with a continuous available time period of 3 or more can be assigned to that node.

[0135] Test nodes are sorted according to their test priority and then assigned to schedulable time slots in descending order of priority. For example, if there are five nodes N1 to N5 to be tested, with test priorities of 0.95, 0.82, 0.76, 0.58, and 0.43 respectively, the sorted order would be N1, N2, N3, N4, and N5. For the highest priority node N1, assuming its test requires two consecutive time windows, the earliest available time slot is first searched within the schedulable time slots. If probe P3 is available in the T1-T5 time slot, N1 can be assigned to P3's T1-T2 time slot. For the next highest priority node N2, assuming its test requires three consecutive time windows, it can be assigned to probe P4's T1-T3 time slot. This process continues until all nodes are assigned or no resources are available.

[0136] When a time slot conflict occurs, the lower-priority test node is postponed to a later available time slot. A time slot conflict means that the time slot required by the current node is already occupied by a higher-priority node. For example, node N3 needs to be tested during the T2-T4 time slot, but probe P5 is only available continuously during the T1-T3 time slot. In this case, it is necessary to find other probes or postpone the test time. If probe P6 is available during the T2-T6 time slot, N3 can be assigned to P6's T2-T4 time slot; if all probes are unavailable during the T2-T4 time slot, N3 needs to be postponed to a later available time slot, such as P5's T5-T7 time slot. The same principle applies to the allocation of lower-priority nodes N4 and N5. If a node still cannot be allocated after multiple postponements, a maximum number of postponements can be set. If this number is exceeded, the node is marked as "unschedulable" and awaits the next round of testing.

[0137] A controllable switch unit is an electronic switch that connects physical probes to test nodes. The connection between a probe and different test nodes is achieved by controlling the opening and closing state of the switch. Assume there are four controllable switch units S1 to S4, each capable of connecting two test nodes. S1 connects nodes N1 and N2, S2 connects nodes N2 and N3, S3 connects nodes N3 and N4, and S4 connects nodes N4 and N5. Based on the probe allocation, a switch control timing sequence is generated. For example, if node N1 is allocated to probe P3 for the T1-T2 time period, then S1 must open channel N1 during the T1-T2 time period; if node N2 is allocated to probe P4 for the T1-T3 time period, then S1 must open channel N2 during the T1-T3 time period, and S2 must also open channel N2 during the T1-T3 time period. The test timing schedule is updated based on the switch control timing sequence to obtain a complete test plan. The test timing schedule includes the operating status of each probe within each time window, the connected test nodes, and the control signals of the switch units. For example, in time window T1, probe P3 connects to node N1, and the N1 channel of switch S1 is opened; probe P4 connects to node N2, and the N2 channels of switches S1 and S2 are opened; the states of the remaining probes and switches remain unchanged. The updated test timing schedule will guide the automated test equipment to perform test operations, ensuring that the tests are conducted in the optimized order.

[0138] This invention achieves optimized allocation of test resources by statistically analyzing probe occupancy status, priority sorting, conflict resolution, and switch control timing, ensuring that high-priority nodes are tested in a timely manner, thereby improving testing efficiency and the detection rate of critical faults.

[0139] Secondly, a multifunctional test point allocation system based on an adaptive switch matrix is ​​provided, including:

[0140] The first unit is used to acquire the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes circuit connection relationships;

[0141] The second unit is used to construct an adaptive switch matrix based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switch units in different time periods, thereby generating a test timing schedule.

[0142] The third unit is used to collect the test response signals of the logic test points, calculate the deviation value of the test response signals relative to the expected signals; determine the fault test points based on the deviation value, analyze the degree of impact on the test nodes connected to the fault test points based on the circuit connection relationship, and obtain the test priority level of the test nodes; adjust the test timing schedule table according to the test priority level, prioritize the allocation of the physical probes to the test nodes with higher priority levels, obtain the updated test timing schedule table, control the switching state of each controllable switch unit, and realize the dynamic allocation of test points.

[0143] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

Claims

1. A multifunctional test point allocation method based on an adaptive switching matrix, characterized in that, include: Obtain the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes circuit connection relationships; An adaptive switch matrix is ​​constructed based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switching unit in different time periods, thereby generating a test timing schedule. The test response signals of the logic test points are collected, and the deviation value of the test response signal relative to the expected signal is calculated. Based on the deviation value, fault test points are determined. The impact on test nodes connected to the fault test points is analyzed based on the circuit connection relationship to obtain the test priority level of the test nodes. Specifically, this includes: constructing a deviation feature vector from the deviation value; standardizing the deviation feature vector and weighting it with a deviation weight coefficient to obtain a fault score; determining fault test points based on the fault score; calculating the conductance coupling coefficient and mutual inductance coupling coefficient between the fault test point and adjacent test nodes; multiplying the conductance coupling coefficient and mutual inductance coupling coefficient by coupling weight coefficients respectively and summing them to obtain the coupling coefficient. The system calculates the first-order influence intensity of the fault test point on adjacent test nodes based on the coupling coefficient matrix; calculates the path attenuation factor based on the connection distance between test nodes, and uses the product of the first-order influence intensity and the path attenuation factor as the propagation influence intensity; calculates the multi-order influence intensity of the fault test point on non-adjacent test nodes based on the propagation influence intensity; multiplies the influence intensities of different orders with the corresponding influence weight coefficients and sums them to obtain the total influence degree of the test node; divides the total influence degree by the maximum total influence degree to obtain the basic priority; calculates the timing adjustment coefficient based on the waiting time of the test node; and uses the product of the basic priority and the timing adjustment coefficient as the test priority level of the test node. Calculate the accuracy of the test priority level, and adjust the deviation weight coefficient, coupling weight coefficient, and influence weight coefficient according to the change in the accuracy; adjust the test timing schedule according to the test priority level, and preferentially allocate the physical probe to the test nodes with higher priority level to obtain the updated test timing schedule, control the switching state of each controllable switch unit, and realize the dynamic allocation of test points.

2. The method according to claim 1, characterized in that, The steps for generating a test timing schedule by controlling the switching state of a controllable switching unit to allocate the same physical probe to different logic test points at different time periods include: Obtain the feature vectors of the logical test points, calculate the correlation matrix between the test points based on the feature vectors of the logical test points, construct the probe allocation state space based on the correlation matrix, the probe allocation state space includes probe allocation state, resource constraints and test quality indicators; construct the probe allocation action space, the probe allocation action space includes probe number, test point number and allocation time period. A reward function is constructed based on the probe allocation state space and the probe allocation action space. The reward function includes a test coverage reward, a test quality reward, and a constraint violation penalty. The probe allocation strategy is iteratively optimized based on the reward function. The probe allocation strategy satisfies the single-moment probe allocation constraint, the test point correlation constraint, and the minimum test time constraint. Collect response signals from test points and calculate test quality assessment values. When the test quality assessment value is lower than a preset assessment threshold, update the weight parameters in the reward function based on the gradient descent method and re-execute the iterative optimization of the probe allocation strategy. Generate a test timing schedule table based on the optimized probe allocation strategy.

3. The method according to claim 2, characterized in that, The probe allocation strategy is iteratively optimized based on the reward function. The steps of the probe allocation strategy satisfying the single-time probe allocation constraint, the test point correlation constraint, and the minimum test time constraint include: A binary decision variable is established for the probe allocation constraint at a single time step. The binary decision variable represents the allocation relationship between the probe and the test point at each time step. The degree of violation of the correlation constraint of the test point is calculated based on the binary decision variable. The degree of violation of each constraint is normalized to obtain the normalized constraint violation degree. The constraint weight is dynamically updated based on the normalized constraint violation degree. The update of the constraint weight is achieved by multiplying the learning rate and the normalized constraint violation degree. A comprehensive evaluation index of constraint satisfaction is calculated based on the constraint weight. A neighborhood structure for probe allocation schemes is constructed, which includes a probe allocation time adjustment scheme and a probe and test point redistribution scheme. Historical schemes are recorded to form a taboo table, and the probe allocation scheme is locally optimized based on the taboo table. The optimization convergence is calculated based on the comprehensive evaluation index, and the optimization step size is dynamically adjusted based on the optimization convergence. The comprehensive evaluation index is used as the optimization basis for the reward function. Calculate the variance of the probe allocation scheme after multiple consecutive optimizations. When the variance is less than a preset variance threshold or the number of consecutive optimizations reaches a preset number of iterations, determine the optimization result of the probe allocation strategy.

4. The method according to claim 1, characterized in that, The steps for calculating the deviation of the test response signal from the expected signal include: The test response signal is filtered to obtain a filtered signal. The filtered signal is divided into signal segments according to a preset time window. An expected signal template is established based on a standard waveform. The root mean square error between the signal segment and the expected signal template is calculated to obtain the time domain deviation. The spectrum between the signal segment and the expected signal template is calculated to obtain the frequency domain deviation. The delay deviation is calculated based on the cross-correlation function between the signal segment and the expected signal template. The time-domain deviation, frequency-domain deviation, and delay deviation are multiplied by their respective weighting coefficients and summed to obtain a comprehensive deviation index. The confidence level of the test response signal is calculated based on the comprehensive deviation index. When the confidence level is lower than a preset confidence threshold, the sampling parameters are adjusted and the test response signal is re-acquired. The comprehensive deviation index is then output as the deviation value of the test response signal relative to the expected signal.

5. The method according to claim 1, characterized in that, The steps of calculating the path attenuation factor based on the connection distance between test nodes, using the product of the first-order influence intensity and the path attenuation factor as the propagation influence intensity, and calculating the multi-order influence intensity of the fault test point on non-adjacent test nodes based on the propagation influence intensity include: Calculate the shortest physical connection path length between test nodes; obtain the frequency characteristics of the test response signal; calculate the frequency-related attenuation coefficient based on the frequency characteristics; use the product of the frequency-related attenuation coefficient and the shortest physical connection path length as the base attenuation value; calculate the branch loss value based on the number of branches in the propagation path; and combine the base attenuation value and the branch loss value to obtain the path attenuation factor. Calculate the equivalent resistance value of the propagation path, determine the path propagation weight based on the equivalent resistance value, and take the product of the first-order influence intensity, the path attenuation factor, and the path propagation weight as the propagation influence intensity. For each non-adjacent test node, the influence intensity propagated from the adjacent node is calculated based on the propagation influence intensity, and the influence intensity propagated from all adjacent nodes is superimposed to obtain the influence intensity of the current order.

6. The method according to claim 1, characterized in that, The step of adjusting the test timing schedule according to the test priority level, and preferentially assigning the physical probes to high-priority test nodes, includes: The occupancy status of probes within each time window is statistically analyzed, and the schedulable time period of physical probes is calculated based on the occupancy status. Test nodes are sorted according to their test priority level, and then assigned to schedulable time periods in descending order of priority. Based on the connection relationship between each controllable switch unit and the test node, the switching control sequence of the controllable switch unit is generated, and the test timing arrangement table is updated based on the switching control sequence.

7. A multifunctional test point allocation system based on an adaptive switch matrix, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to acquire the circuit topology information of the integrated circuit under test, wherein the circuit topology information includes circuit connection relationships; The second unit is used to construct an adaptive switch matrix based on the circuit topology information. The adaptive switch matrix includes multiple controllable switch units for connecting physical probes and logic test points. Based on time-division multiplexing, the same physical probe is assigned to different logic test points by controlling the switching state of the controllable switching unit in different time periods, thereby generating a test timing schedule. The third unit is used to collect the test response signal of the logic test point, calculate the deviation value of the test response signal relative to the expected signal; determine the fault test point based on the deviation value, analyze the degree of impact on the test nodes connected to the fault test point based on the circuit connection relationship, and obtain the test priority level of the test node. The test timing schedule is adjusted according to the test priority level, and the physical probes are preferentially assigned to test nodes with higher priority levels to obtain an updated test timing schedule. The switching state of each controllable switch unit is controlled to realize the dynamic allocation of test points.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images